| 1 |
On Oct 14, 2013 6:04 PM, "Tanstaafl" <tanstaafl@×××××××××××.org> wrote: |
| 2 |
> |
| 3 |
> On 2013-10-13 5:49 PM, Dale <rdalek1967@×××××.com> wrote: |
| 4 |
>> |
| 5 |
>> Talk about putting some stuff on tmpfs. O_O I have always wanted to |
| 6 |
>> copy the tree to tmpfs and run "time emerge -uvaDN world". Just to see |
| 7 |
>> how fast it will go. lol |
| 8 |
> |
| 9 |
> |
| 10 |
> I remember once I worked for an Apple reseller that had this accounting |
| 11 |
program that required them to do some kind of 'reconciliation' every month |
| 12 |
that required a massive amount of processing - it took like 36 hours or |
| 13 |
something ridiculous (literally almost took all weekend), and he had |
| 14 |
implemented a rule that someone had to be there the entire time to baby sit |
| 15 |
the process - apparently it wasn't uncommon for there to be an error that |
| 16 |
would require them to restart it - and this was on a pretty powerful system |
| 17 |
at the time. |
| 18 |
> |
| 19 |
> Well, one weekend, when we were building a system for a customer with |
| 20 |
tons of RAM (for the time) I talked them into a little experiment. The boss |
| 21 |
didn't believe me when I told him I could get the reconciliation processing |
| 22 |
time down to less than a day (I told him probably just a few hours, but |
| 23 |
wasn't sure)... so we made a bet. |
| 24 |
> |
| 25 |
> I took a Quadra 900 (or maybe it was a 950), and added a bunch of RAM - I |
| 26 |
think we got it up to 128MB or something ridiculous (this was in about |
| 27 |
1992). The accounting DB was about 40MB at the time, but hey, we had the |
| 28 |
RAM, so I just loaded it up. |
| 29 |
> |
| 30 |
> I created a RAM disk, copied the entire Accounting DB into it, and |
| 31 |
started running the reconciliation. The process finished after about 45 |
| 32 |
minutes (I was even surprised at that), and while there were no errors and |
| 33 |
it said it had completed successfully, the boss was sure that something had |
| 34 |
gone wrong. So, he re-ran it the old way on the old server, and almost 2 |
| 35 |
days later, when the numbers matched, he just shook his head and paid me |
| 36 |
off, muttering about the lost weekends over the last 5 years he'd been |
| 37 |
there. He kept that machine around for running the reconciliation for at |
| 38 |
least a few months, but then I left, so no idea how long he kept it for... |
| 39 |
> |
| 40 |
|
| 41 |
Niiiice.. 48x performance improvement? I know of some DBA who would gladly |
| 42 |
pay an arm + a leg + their grandmothers for that kind of improvement :-) |
| 43 |
|
| 44 |
Kind of tangential, but that's what Oracle is aiming with their TimesTen |
| 45 |
product: give the server oodles of RAM, and load the database in memory. |
| 46 |
|
| 47 |
Another similar performance-improving method would be using Fusion-IO to |
| 48 |
load the database into direct-memory-mapped SSDs. They claimed that the |
| 49 |
most high-end Fusion-IO devices can reach up to 9 million IOPS... |
| 50 |
|
| 51 |
Rgds, |
| 52 |
-- |
Archives